I build agentic systems for games and study how worlds can remember. Founder of WorldLines — an agentic engine for games with Persistent Narrative World Representation, where AI-driven storytelling meets interactive game worlds. Ph.D. in Computer Graphics at the University of Tokyo (advised by Prof. Takashi Kanai), focusing on physics-based animation and multi-phase SDF representations for fracture. Previously Game Engineer at DeNA and IoT full-stack at Butlr.
Agentic games, research systems, and interactive demos — newest first.
An agentic engine for games with Persistent Narrative World Representation — a world that records and recalls what the player does. Agents play characters, the world keeps state, and every choice is remembered. The platform turns a script into a living world you can branch, restore, and share.
A living world running on WorldLines — every character is an independent soul-agent with its own mind, memory, and agenda. Relationships form on the soul graph, stories accumulate into eras, and the world keeps living whether you watch or not.
A validation slice testing whether an LLM can run a fair game master: dice-gated checks, persistent party state, and consequences that carry across a long descent. A playable answer to "can agents hold a tabletop expedition together?" — built on the WorldLines engine.
A swipe-native, short-form branching RPG built as a top-of-funnel format — measuring how far players go and whether they convert. A dice-gated story you play one swipe at a time, designed to bring new players into the WorldLines world.
A generative approach for predicting brittle fractures with neural discrete representation learning, published in Computer Graphics Forum (2025). Awarded the GRSI Replicability Stamp for reproducible research.
An interactive demo and dataset for brittle-fracture prediction — run the DeepFracture models directly on HuggingFace Spaces, and explore the underlying break4models dataset.
A spatial analytics and visualization metrics system built during the DeNA game-engineering years — turning play-space data into legible visual metrics for design decisions.
An automated voting pipeline for the AKB Sousenkyo — an early hobby project in Python automation and web scraping, and a first taste of building systems that act on their own.
Writing from the lab, plus conference talks and media.
My work sits between games, AI agents, and computer graphics research. Some of it is published research on physics-based animation; some is shipping, playable systems where autonomous agents drive narrative and play. WorldLines is where those two threads meet — an attempt to build worlds that remember.
If we are standing at the early stage of agent-native games, I want to see that landscape clearly — and build the tools that let more people make worlds of their own.
Presented a work at Eurographics 2026 (EG 2026).
